Re: [PATCH v2 15/17] libnvdimm: Set numa_node to NVDIMM devices

[Dan Williams <dan.j.williams@intel.com>]

On Thu, Jun 25, 2015 at 11:34 AM, Williams, Dan J
<dan.j.williams@intel.com> wrote:
> On Thu, 2015-06-25 at 11:45 -0600, Toshi Kani wrote:
>> On Thu, 2015-06-25 at 05:37 -0400, Dan Williams wrote:
>> > From: Toshi Kani <toshi.kani@hp.com>
>> >
>> > ACPI NFIT table has System Physical Address Range Structure entries that
>> > describe a proximity ID of each range when ACPI_NFIT_PROXIMITY_VALID is
>> > set in the flags.
>> >
>> > Change acpi_nfit_register_region() to map a proximity ID to its node ID,
>> > and set it to a new numa_node field of nd_region_desc, which is then
>> > conveyed to the nd_region device.
>> >
>> > The device core arranges for btt and namespace devices to inherit their
>> > node from their parent region.
>> >
>> > Signed-off-by: Toshi Kani <toshi.kani@hp.com>
>> > [djbw: move set_dev_node() from region 'probe' to 'create']
>>
>> Sorry, I failed to mention other issue, which led me call set_dev_node()
>> in probe.  nd_async_device_register() calls device_add(), which does:
>>
>>         /* use parent numa_node */
>>         if (parent)
>>                 set_dev_node(dev, dev_to_node(parent));
>>
>> and overwrites numa_node to -1.  Since region's parent is ndbusN, we
>> cannot set numa_node to the parent.  So, I had to set it in probe.
>
> In general, I still don't like leaving it up to ->probe() which is
> within its rights to fail and not set the node.  How about the following
> that moves it to the bus uevent code?  Should get triggered before probe
> so the numa_node is valid before userspace is ever notified about the
> device.
>
> device_add() does:
>
>         kobject_uevent(&dev->kobj, KOBJ_ADD);
>         bus_probe_device(dev);
>
> ...so I think we're good, agree?  I also added a missing init of
> ndr_desc.numa_node in arch/x86/kernel/pmem.c, see below.

This looks good in a quick manual test.  It's interesting/illustrative
that I inadvertently broke the one bit of the libnvdimm sysfs
interface that did not have unit test coverage.